lambda128
Download

v0.2.0 open source desktop ready

lambda128

Agentic coding that stays in your editor.

A local-first AI pair programmer for VS Code and VS Codium — built on Code-OSS, inspired by Cursor's workflow. Chat, plan, edit, and approve every change. Bring your own keys.

Download v0.2.0 View on GitHub

Download

Desktop binaries are here.

v0.2.0 ships as a standalone desktop app for Linux, macOS, and Windows. Download, make executable, and run — no build step required.

Linux AppImage 123 MB Linux .deb 85 MB macOS .dmg 113 MB Windows .exe 93 MB

Linux: chmod +x lambda128-0.2.0.AppImage && ./lambda128-0.2.0.AppImage
All releases at github.com/androvonx95/Lambda128/releases

Current status

v0.2.0 — desktop apps shipping.

lambda128 is now available as a standalone desktop application for Linux, macOS, and Windows. 39 tests passing, 0 TypeScript errors, CI/CD pipeline with automatic releases on every tag.

What works

  • Chat with streaming responses
  • Agent mode — plan, execute, observe, replan
  • 11 tools (read, write, edit, search, git, terminal)
  • 6 providers + automatic failover
  • Local Ollama models (llama3.2, codellama, qwen2.5-coder…)
  • Patch/diff review before any file is written
  • Permission tiers — auto-approve reads, ask for writes
  • Conversation history in local SQLite
  • Desktop app packaging (.AppImage / .deb / .dmg / .exe)
  • CI/CD with automatic GitHub Releases

What's coming

  • VS Code Marketplace & Open VSX publishing
  • Tab autocomplete (FIM)
  • Semantic search UI (embeddings engine built)
  • Parallel tool execution for faster agent loops

Workflow

Plan. Read. Edit. Approve.

lambda128 mirrors a Cursor-inspired agent loop — without leaving the editor you already use, and without a hosted backend.

  1. 01

    Ask or launch an agent

    Chat about your codebase, or prefix with @agent for multi-step autonomous work.

  2. 02

    Tools explore the workspace

    The agent reads files, searches with ripgrep, checks git status/diffs, and runs terminal commands — all within your project boundary.

  3. 03

    You approve the diffs

    Edits arrive as SEARCH/REPLACE patches. Nothing lands on disk until you say yes. Terminal commands and file deletes require explicit approval.

Terminal — lambda128
$ git clone https://github.com/androvonx95/Lambda128.git Cloning into 'Lambda128'... done. $ cd Lambda128 && pnpm install && pnpm -r build Packages: 8 built in 12.4s $ node test-caches.mjs ✓ 39 passed, 0 failed $ cd packages/desktop && npx electron-builder --linux ✓ lambda128-0.2.0.AppImage (123 MB) $ ▊

Capabilities

Built for real coding sessions.

Context-aware chat, an agent with a full toolbelt, inline actions on selected code, and automatic provider failover when a model goes down.

Chat mode

Open the AI panel. The model sees your open file, selection, project structure, and git status.

Agent mode

Decompose a task, search the tree, generate edits, and pause for approval before writing.

Inline actions

Select code → Explain, Fix, Refactor from the context menu or Command Palette.

Provider failover

If Anthropic goes down, lambda128 routes to OpenAI — or whichever provider you configure next.

Token economy

File cache, context dedup, and automatic history trimming keep prompts lean within model windows.

Local-first keys

API keys live in your OS keychain — macOS Keychain, Windows Credential Manager, Linux Secret Service.

Architecture

Eight-package monorepo.

TypeScript throughout. Local-first design with a React chat webview, Node.js agent engine, and SQLite storage — no cloud backend required.

shared

Types, constants, IPC contracts

core

Agent engine, prompt orchestrator, tool registry, token budget

providers

OpenAI, Anthropic, Gemini, OpenRouter, Bedrock, Ollama

storage

SQLite conversations, OS keychain, file cache

repository

Workspace scanner, repo-map, embeddings engine

vscode-extension

Chat webview, settings, history, agent UI

webview

React chat UI with streaming & diff viewer

desktop

Electron desktop app shell

Full 16-section design document in ARCHITECTURE.md on GitHub.

Security

Your machine. Your rules.

Every provider call goes directly from your machine. The AI cannot modify files, run commands, or read outside your workspace without your consent.

OS keychain storage

API keys never sit in plaintext settings or get committed to git.

No cloud backend

No product account, no telemetry pipeline, no hosted inference layer.

Approval-gated writes

File edits, terminal commands, and deletes all require explicit approval.

Workspace boundary

The agent cannot read files outside your open project folder.

Build from source

Prefer to build it yourself?

Requirements: Node.js 22+, pnpm 11+.

1

Clone the repository

bash 
git clone https://github.com/androvonx95/Lambda128.git
cd Lambda128
2

Install dependencies & build

bash 
pnpm install
pnpm -r build
3

Run the test suite

bash 
node test-caches.mjs
# Expected: 39 passed, 0 failed

Run

Launch inside VS Code or VS Codium.

Open the monorepo in your editor, then start the Extension Development Host.

  1. Open the project File → Open Folder… and select the repository root.
  2. Start debugging Press F5 (or Run → Start Debugging) to open a new Extension Development Host window.
  3. Open the AI panel In the new window, click the activity-bar AI Assistant icon, or run lambda128: Open Chat from the Command Palette.
  4. Package a VSIX (optional) From packages/vscode-extension, run pnpm package for a distributable .vsix.
bash — package extension 
cd packages/vscode-extension
pnpm package

Configure

Bring your own API keys.

Open Settings (Ctrl+, / Cmd+,) and search for lambda128. Configure at least one provider. Keys are stored in your OS keychain.

Setting Purpose
lambda128.anthropicApiKey Anthropic API key (recommended default)
lambda128.openaiApiKey OpenAI API key
lambda128.geminiApiKey Google Gemini API key
lambda128.openrouterApiKey OpenRouter API key
lambda128.bedrockAccessKeyId AWS Bedrock access key
lambda128.ollamaModel Local Ollama model (default: llama3.2)
lambda128.ollamaBaseUrl Ollama API URL (default: http://localhost:11434/v1)
lambda128.defaultProvider Default provider (e.g. anthropic)
lambda128.embeddingMode off · local · cloud

Use

Day-to-day commands.

Once the extension is running and a key is set, these are the moves you'll use most.

Chat

Open the Chat view in the AI Assistant sidebar. Ask about the open file, a selection, or the wider project.

Placeholder: Ask anything about your code… (Use @agent for agent mode)

Start agent mode

Type @agent in chat, or run lambda128: Start Agent Mode from the Command Palette.

  • Plans the task into steps
  • Reads and searches relevant files
  • Produces diffs and asks before writing

Inline on a selection

Highlight code, then right-click or use the Command Palette:

  • lambda128: Explain This Code
  • lambda128: Fix This Code
  • lambda128: Refactor This Code

Agent tools

Tool Approval
read_fileAuto
search_files · glob · list_directoryAuto
git_status · git_diffAuto
write_file · edit_fileRequired
run_terminalRequired
delete_file · rename_fileRequired

Providers

Six providers. Automatic failover.

Pick a model family — cloud or local — or let failover route around an outage.

  • OpenAI GPT-4o · GPT-4o-mini
  • Anthropic Claude Sonnet · Haiku
  • Google Gemini gemini-2.5-flash & friends
  • OpenRouter route to many models
  • AWS Bedrock Claude · Llama · Nova
  • Ollama llama3.2 · codellama · qwen2.5-coder · deepseek-coder · codestral

lambda128

Your editor. Your keys. Your approval.

Download the desktop app, drop in an API key, and start an agent.

Download v0.2.0 Star on GitHub

Get notified at launch

No spam. One email when lambda128 hits the marketplace.